1: % alf.nbi.dk:~predrag/articles/hongkong/hk.tex
2: % Predrag 5 nov 1999
3: % Predrag 29 oct 1999
4: % Predrag 15 oct 1999
5:
6: %
7: % the paper itself starts with line *** START ***
8: %
9:
10:
11: \documentclass{elsart}
12: %\documentclass{article}
13: %\usepackage{natbib}
14: \usepackage{graphicx}
15: %\usepackage[draft]{graphicx} %% draft mode for figures
16: \usepackage{ifthen} %% used for boolean logic commands
17: %\usepackage{amssymb}
18:
19:
20: \input defHK
21:
22: %************ START ***************************************
23: \begin{document}
24: \runauthor{Cvitanovi\c'}
25: \begin{frontmatter}
26: \title{Chaotic Field Theory: a Sketch}
27: %Unstable Recurrent Patterns and
28: %Nonlinear Field Transformations
29: %Semiclassical Field Theory}
30: \author{Predrag Cvitanovi\'c} %\thanksref{Version}}
31: % \author[Predrag]{Predrag Cvitanovi\'c\thanksref{Someone}}
32: % \thanks[Version]{
33: % Nov 4, 1999 draft $\sim$/articles/hongkong/hk.tex
34: % }
35:
36: \address{
37: %\address[Predrag]{
38: Department of Physics \&\ Astronomy, Northwestern University \\
39: 2145 Sheridan Road, Evanston, Illinois 60208
40: }
41:
42: %\thanks[Someone]{Partially supported by the Roman Senate}
43:
44: \begin{abstract}
45: Spatio-temporally chaotic dynamics of a
46: classical field can be described
47: by means of an infinite hierarchy of its unstable spatio-temporally
48: periodic solutions.
49: The periodic orbit theory yields the
50: global averages characterizing the chaotic dynamics, as well
51: as the starting semiclassical approximation to the quantum theory.
52:
53: New methods for computing corrections to the semiclassical
54: approximation are developed; in particular, a nonlinear field
55: transformation yields
56: the perturbative corrections in a form more compact than
57: the Feynman diagram expansions.
58: \end{abstract}
59:
60: \begin{keyword}
61: Periodic orbits, field theory, semiclassical quantization,
62: trace formulas
63: \\
64: PACS {02.50.Ey, 03.20.+i, 03.65.Sq, 05.40.+j, 05.45.+b}
65: \end{keyword}
66: \end{frontmatter}
67:
68: %\section{Introduction}
69:
70: Formulated in 1946-49 and tested through 1970's, quantum electrodynamics takes
71: free electrons and photons as its point of departure,
72: % (``harmonic oscillators'',
73: % ``%integrable'' 0-th order theory),
74: with nonlinear effects taken
75: in account perturbatively in terms of Feynman diagrams, as corrections
76: of order $({\alpha / \pi})^n = (0.002322819\dots)^n$.
77: % 3.14159265358979323844
78: % 1/alpha = 137.035 999
79: % \alpha / \pi = .00232281946719555546
80: QED is a wildly successful theory, with Kinoshita's\rf{KinoHughes}
81: calculation of the electron magnetic moment
82: \[
83: {1 \over 2} (g-2) = \sum_n \left({\alpha \over \pi}\right)^n
84: \left\{ \cdots \,+\, \btrackA{QEDgraph} \,+\,\cdots \right\}
85: %\,.
86: \]
87: % \bea
88: % a_e(theo) &=&
89: % {1 \over 2} (g-2) = {1 \over 2} {\alpha \over \pi}
90: % - 0.32848 \left({\alpha \over \pi}\right)^2
91: % + (1.183 \pm 0.011) \left({\alpha \over \pi}\right)^3.
92: % \continue
93: %&=& 1 159 652 ???.4 (4.3) \times 10^{-12}
94: %\nnu
95: %\eea
96: agreeing with Dehmelt's experiments\rf{Dehmelt87} to
97: 12 significant digits.
98: % unparalleled precision
99: % \[
100: % a_e(expt) = 1 159 652 188.4 (4.3) \times 10^{-12}
101: % \]
102:
103: Quantum chromodynamics perturbative calculations
104: seemed the natural next step,
105: the only new feature being the gluon-gluon interactions.
106: However, in this case the Feynman--diagrammatic expansions
107: for observables such as the meson and hadron masses
108: \[
109: \mbox{(observable)} = \sum_n \left(\alpha_{QCD}\right)^n
110: \left\{ \cdots \,+\, \btrackA{QCDgraph} \,+\,\cdots \right\}
111: \]
112: failed us utterly, perhaps because
113: % (with the exception of the deep inelastic regime)
114: the expansion parameter is of order 1. I say perhaps,
115: because more likely the error in
116: this case is thinking in terms of quarks and gluons
117: % (``harmonic oscillators'', ``integrable'' 0-th order theory),
118: in the first place. Strongly nonlinear field theories
119: require radically different approaches,
120: and in 1970's, with a deeper appreciation of the connections between
121: field theory and statistical mechanics, their re-examination led
122: to path integral formulations such as the lattice QCD\rf{KWilson}.
123: In lattice theories quantum fluctuations explore the full gauge group
124: manifold, and classical dynamics of Yang-Mills fields plays no role.
125:
126: We propose to re-examine here the path integral formulation
127: and the role that the classical solutions play
128: in quantization of strongly nonlinear fields.
129: In the path integral formulation of a field theory the dominant
130: contributions come from saddlepoints, the classical solutions
131: of equations of motion. Usually one imagines one
132: dominant saddle point, the ``vacuum'':
133: \vspace{2ex}
134: \\
135: {\centerline{
136: ${
137: \hspace{-4ex}
138: \includegraphics[width=0.30\textwidth]{FTfig3-14.eps}
139: \hspace{-4ex}
140: \atop
141: \mbox{one dominant extremum}
142: }$
143: ~~~~
144: ${
145: \hspace{-2ex}
146: %\raisebox{4.0ex}[5.5ex][4.5ex]
147: {
148: \includegraphics[width=0.28\textwidth]{instanton.eps}
149: }
150: %\vspace{4ex}
151: \hspace{-2ex}
152: \atop
153: %\vspace{4ex}
154: \mbox{an infinity of instanton saddles}
155: }$
156: }}
157: \vspace{1ex}
158: \\
159: The Feynman diagrams of QED and QCD are
160: nothing more than a scheme to compute the correction terms to
161: this starting
162: semiclassical, Gaussian saddlepoint approximation.
163: But there might be other saddles.
164: That field theories might have
165: a rich repertoire of classical solutions became apparent
166: with the discovery of instantons\rf{Polyakov}, analytic
167: solutions of the classical $SU(2)$ Yang-Mills equations of
168: motion, and the realization that the associated instanton vacua
169: receive contributions from countable $\infty$'s of
170: saddles. What is not clear is whether these are the important
171: classical saddles. Could it be that the strongly nonlinear theories
172: are dominated by altogether different
173: classical solutions?
174:
175: The search for the classical
176: solutions of nonlinear field theories such as the Yang-Mills and
177: gravity has so far been neither very successful nor very systematic.
178: % In part because nobody is looking.
179: In modern field theories the main emphasis has been
180: on symmetries
181: %(compactly collected in action functionals that define the theories)
182: as guiding principles in writing down the actions.
183: % The infinitesimal time, classical
184: % equations of motion follow by variational principles.
185: But writing down a differential equation is only the start of the story;
186: even for systems as simple as
187: 3 coupled ordinary differential equations one in general
188: has no clue what the nature of the long time solutions might be.
189:
190: These are hard problems, and
191: in explorations of modern field theories the dynamics tends to
192: be is neglected, and
193: understandably so, because the wealth of the classical solutions
194: of nonlinear systems can be truly bewildering.
195: If the classical behavior of these theories is anything like
196: that of
197: the field theories that describe the classical world ---
198: the hydrodynamics, the
199: magneto-hydrodynamics, the
200: % Burgers,
201: Ginzburg-Landau
202: % Kuramoto-Sivashinsky
203: system
204: ---
205: there should be very many solutions, with very few of the important
206: ones analytical in form; the strongly nonlinear classical field theories
207: are turbulent, after all.
208: Furthermore, there is not a dimmest hope that such solutions
209: are either beautiful or analytic,
210: and there is not much enthusiasm for grinding out
211: numerical solutions
212: as long as one lacks ideas as what to do with them.
213:
214: By late 1970's it was generally
215: understood that even the simplest nonlinear systems
216: exhibit chaos. Chaos is the norm also for generic
217: Hamiltonian flows, and for path integrals that implies that
218: instead of a few, or countably few saddles,
219: classical solutions populate
220: fractal sets of saddles.
221: \vspace{1ex}
222: \\
223: \centerline{
224: ${
225: \hspace{-4ex}
226: \includegraphics[width=0.40\textwidth]{saddle.eps}
227: \hspace{-4ex}
228: \atop
229: \mbox{a local unstable extremum}
230: }$
231: ~~~~~~~~
232: ${
233: \includegraphics[width=0.35\textwidth]{mountain-c.eps}
234: \atop
235: \mbox{a fractal set of saddles}
236: }$
237: }
238: \vspace{1ex}
239: \\
240: For the path-integral formulation of quantum mechanics
241: such solutions
242: were discovered and accounted for by Gutzwiller\rf{gutbook}
243: in late 1960's.
244: In this framework the spectrum of the theory is computed from
245: a set of its unstable classical periodic solutions.
246: % and quantum corrections.
247: The new aspect is that the individual saddles for classically
248: chaotic systems are nothing like the harmonic oscillator degrees
249: of freedom, the quarks and gluons of QCD --- they are all
250: unstable and highly nontrivial, accessible only by numerical techniques.
251:
252:
253: So, if one is to develop a semiclassical field theory of systems that
254: are classically chaotic or ``turbulent'',
255: the problem one faces is twofold
256: \begin{enumerate}
257: \item
258: determine, classify, and order by relative importance
259: the classical solutions of nonlinear field theories.
260:
261: \item
262: develop methods for calculating perturbative corrections
263: to the corresponding classical saddles.
264:
265: \end{enumerate}
266:
267: Our purpose here is to give an overview over the status of
268: this program --- for details the reader is referred to
269: the literature cited.
270:
271: The first task, a systematic exploration of solutions
272: of field theory has so far been implemented only for one of the very
273: simplest field theories, the 1-dimensional Kuramoto-Sivashinsky system.
274: We sketch below how its spatio-temporally chaotic dynamics can be
275: described in terms of spatio-temporally recurrent
276: unstable patterns.
277:
278: For the second task, the theory of perturbative corrections, we shall turn to
279: an even simpler system; a weakly stochastic mapping
280: in 1-dimension.
281: The new aspect of the theory is that now
282: the corrections have to be computed saddle by saddle.
283: In \refsect{s-PerCorr} to \refsect{scfpo} we discuss
284: three distinct methods for their evaluation.
285:
286:
287: \section{Unstable recurrent patterns in classical field theories}
288:
289: Field theories such as
290: 4-dimensional QCD or gravity have many dimensions, symmetries,
291: tensorial indices. They are
292: far too complicated for exploratory forays into
293: this forbidding terrain.
294: We start instead by taking a simple spatio-temporally chaotic
295: nonlinear system of physical interest, and investigate the
296: nature of its solutions.
297: % very much in the spirit of
298: % examining solutions of the simplest nonlinear 1-dimensional
299: % system, the logistic map $ x \to f(x)=A x (1-x)$.
300:
301: %\subsection{Kuramoto-Sivashinsky system}
302:
303: One of the simplest and extensively studied
304: spatially extended dynamical systems is
305: the Kuramoto-Sivashinsky system\rf{KurSiv}
306: \beq
307: u_t=(u^2)_x-u_{xx}-\nu u_{xxxx}
308: \ee{ks}
309: which arises as an amplitude equation for interfacial instabilities
310: in a variety of contexts.
311: The ``flame front'' $u(x,t)$ has compact support, with
312: $x \in [0,2\pi]$ a periodic space coordinate.
313: The $u^2$ term makes this a nonlinear system,
314: $t$
315: %$t \geq 0$
316: is the time,
317: and $\nu$ is a fourth-order ``viscosity'' damping parameter
318: that irons out any sharp features.
319: % The subscripts $x$ and $t$ denote the partial derivatives with respect to
320: % $x$ and $t$.
321: Numerical simulations demonstrate that as
322: the viscosity decreases (or the size of the system increases),
323: the ``flame front'' becomes increasingly unstable and turbulent.
324: The task of the theory is to describe this spatio-temporal
325: turbulence and yield quantitative predictions for its measurable
326: consequences.
327:
328: Armed with a computer and a great deal of skill,
329: one can obtain a numerical solution to a nonlinear
330: PDE. The real question is;
331: once a solution is found, what is to be done with it?
332: The periodic orbit theory is an answer to this question.
333:
334: Dynamics drives a given spatially extended system through a
335: repertoire of unstable patterns; as we watch
336: a ``turbulent'' system evolve,
337: every so often we catch a glimpse of a familiar pattern:
338: \vspace{4ex}
339: \\
340: \centerline{
341: \raisebox{-4.0ex}[5.5ex][4.5ex]
342: {
343: \includegraphics[height=12ex]{Hopf-a.eps}
344: }
345: ~~~
346: $\Longrightarrow$
347: ~~
348: {other swirls}
349: ~~
350: $\Longrightarrow$
351: ~~~
352: \raisebox{-4.0ex}[5.5ex][4.5ex]
353: {
354: \includegraphics[height=12ex]{Hopf-b.eps}
355: }
356: }
357: \vspace{1ex}
358: \\
359: For any finite spatial resolution,
360: the system follows approximately for a finite time
361: a pattern belonging to a finite
362: alphabet of admissible patterns, and the long term dynamics can be thought
363: of as a walk through the space of such patterns,
364: just as chaotic dynamics with a low dimensional
365: attractor can be thought of as a succession of nearly periodic (but
366: unstable) motions.
367: The periodic orbit provides the
368: machinery that converts this intuitive picture into
369: precise calculation scheme that extracts asymptotic time predictions
370: from the short time dynamics.
371: For extended systems the theory
372: gives a description of the asymptotics of
373: partial differential equations
374: in terms of recurrent spatio-temporal patterns.
375:
376: Putkaradze has proposed that
377: the Kuramoto-Sivashinsky system \refeq{ks}
378: be used as a laboratory for exploring such ideas.
379: We now summarize the results obtained so far in this direction
380: by
381: Christiansen et al.\rf{CCP96} and
382: Zoldi and Greenside\rf{ZG96}.
383: % \PC{get the parameter from Ronnie}
384:
385:
386: The solution $u(x,t)=u(x+2\pi,t)$ is periodic on
387: the $x \in [0,2\pi]$ interval, so one (but by no means only)
388: way to solve such equations is
389: to expand $u(x,t)$ in a discrete spatial Fourier series
390: \begin{equation}
391: u(x,t)= i \sum_{k=-\infty}^{+ \infty} a_k(t) e^{i k x}
392: \, .
393: \label{fseries}
394: \end{equation}
395: Restrict the consideration
396: to the subspace of odd solutions $u(x,t)=-u(-x,t)$ for which
397: $a_k$ are real.
398: Substitution of (\ref{fseries}) into (\ref{ks}) yields
399: the infinite ladder of evolution equations for the Fourier coefficients $a_k$:
400: \begin{equation}
401: \dot{a}_k=(k^2-\nu k^4)a_k - k \sum_{m=-\infty}^{\infty} a_m a_{k-m}
402: \,.
403: \label{expan}
404: \end{equation}
405: $u(x,t)=0$ is a fixed point of (\ref{ks}), with
406: % From (\ref{expan}) it follows that
407: the $k^2 \nu<1$
408: long wavelength modes of this fixed point
409: linearly unstable, and the
410: short wavelength modes stable.
411: For $\nu > 1$, $u(x,t)=0$ is the globally attractive stable fixed point;
412: starting with $\nu =1$ the solutions go through a rich sequence of
413: bifurcations,
414: % whose detailed knowledge is not needed for purpose at hand.
415: and myriad unstable periodic solutions whose number
416: grows exponentially with time.
417:
418: The essential limitation on
419: the numerical studies undertaken so far have been
420: computational constraints: in
421: truncation of high modes in the expansion (\ref{expan}),
422: sufficiently many have to be retained to ensure
423: the dynamics is accurately represented.
424: Christiansen et al.\rf{CCP96} have
425: examined the dynamics for values of the damping parameter
426: %such as $\nu=0.029910$, $\nu=0.029924$,
427: close to the onset of chaos, while
428: Zoldi and Greenside\rf{ZG96}
429: have explored somewhat more turbulent values of $\nu$.
430: With improvement of numerical codes
431: considerably more turbulent regimes should become accessible.
432:
433: One pleasant surprise is that
434: even though one is dealing with (infinite
435: dimensional) PDEs, for these
436: strong dissipation values of parameters the
437: spatio-temporal chaos is sufficiently weak that the
438: flow can be visualised as an approximately 1-dimensional Poincar\'e return map
439: $s \rightarrow f(s)$
440: from the unstable manifold of the shortest periodic point
441: onto its neighborhood, see \reffig{unfolded}(a).
442: This representation makes it possible to
443: systematically determine all nearby periodic solutions up to a given
444: maximal period.
445: %
446: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
447: \FIG{
448: (a)
449: \hspace{-4ex}
450: \includegraphics[width=0.48\textwidth]{unfolded.ps}
451: \hspace{-4ex}
452: ~~~~~~~~
453: (b)
454: \hspace{-4ex}
455: \includegraphics[width=0.48\textwidth]{eigenvalues.ps}
456: \hspace{-4ex}
457: }{}{
458: (a) The return map $s_{n+1} = f(s_n)$ constructed
459: from periodic solutions of the Kuramoto-Sivashinsky
460: equations \refeq{ks}, $\nu=0.029910$,
461: with $s$ the distance
462: measured along the unstable manifold of the fixed
463: point $\overline{1}$.
464: Periodic points $\overline{0}$
465: and $\overline{01}$ are also indicated.
466: (b) Lyapunov exponents $\lambda_k$ versus $k$ for the
467: periodic orbit $\overline{1}$ compared with the stability eigenvalues
468: of the $u(x,t)=0$ stationary solution $k^2- \nu k^4$.
469: $\lambda_k$ for $k \geq 8$ lie below the numerical accuracy of integration
470: and are not meaningful.
471: From \refref{CCP96}.
472: }{unfolded}
473: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
474: %
475:
476: So far some 1,000 prime cycles have been determined numerically
477: for various values of viscosity. %\rf{CCP96,ZG96}.
478: The rapid contraction in the nonleading eigendirections
479: is illustrated in \reffig{unfolded}(b)
480: by the plot of the first 16 eigenvalues of the
481: $\overline{1}$-cycle.
482: As the length of the orbit increases, the magnitude of
483: contracting eigenvalues falls off very quickly.
484: In \reffig{orbit0fig} we plot
485: $u_0(x,t)$ corresponding to the $\overline{0}$-cycle.
486: %%%% VP replaced this 26/6-96
487: % As typical differences between different solutions are
488: % considerably smaller than the overall amplitudes,
489: %%%% by this:
490: The difference between this solution and the other shortest period
491: solution is of the order of 50\% of a typical variation in
492: the amplitude of $u(x,t)$, so the chaotic dynamics is
493: already exploring a sizable swath in the space of possible
494: patterns even so close to the onset of spatio-temporal chaos.
495: Other solutions, plotted in the configuration space, exhibit the same
496: overall gross structure.
497: Together they form the
498: repertoire of the recurrent spatio-temporal patterns that is being
499: explored by the turbulent dynamics.
500:
501: \FIG{
502: (a)
503: \hspace{-4ex}
504: \includegraphics[width=0.50\textwidth]{orbit0.ps}
505: \hspace{-4ex}
506: (b)
507: \hspace{-4ex}
508: \includegraphics[width=0.50\textwidth]{diff1.ps}
509: \hspace{-4ex}
510: }{}{
511: (a) Spatio-temporally periodic solution $u_0(x,t)$ of the
512: Kuramoto-Sivashinsky system,
513: viscosity parameter $\nu=0.029910$.
514: (b) The difference between the two shortest period
515: spatio-temporally periodic solutions
516: $u_0(x,t\period{0})$ and $u_1(x,t\period{1})$.
517: From \refref{CCP96}.
518: }{orbit0fig}
519:
520: \section{Periodic orbit theory}
521: \label{PerOrsInAction}
522:
523: Now we turn to the central issue; qualitatively, these
524: solutions demonstrate that the recurrent patterns program
525: can be implemented, but how
526: is this information to be used quantitatively? This is
527: what the periodic orbit theory is about;
528: it offers the machinery that assembles
529: the topological and the quantitative information about individual
530: solutions
531: % such as their periods and stabilities,
532: into accurate predictions
533: about measurable global averages, such as the Lyapunov exponents
534: and correlation functions.
535:
536: Very briefly (for a detailed exposition
537: the reader is referred to \refref{QCcourse}),
538: the task of any theory that aspires to be a theory of
539: chaotic, turbulent systems is is to predict the value
540: of an ``observable" $\obser$
541: from the spatial and time averages evaluated along
542: dynamical trajectories $x(t)$
543: % = f^t(x)$
544: \[
545: \left<\obser\right> = \lim_{t\to\infty} {1 \over t} \left<\Obser^t \right>
546: \,,\qquad
547: \Obser^t (x) = \int_0^t \d\tau \, \obser(x(\tau))
548: \,.
549: \]
550: The key idea of the periodic orbit theory
551: is to extract this average from the leading eigenvalue
552: % $\e^{ts(\beta)}$
553: of the {\evOper}
554: \[
555: {\Lop}^t (x,y) = \delta(y-x(t))\e^{\beta \Obser^t(x)}
556: \]
557: via the trace formula
558: \beq
559: \tr {\Lop}^{t}
560: = \sum_{p}
561: \left.
562: \raisebox{-4.0ex}[5.5ex][4.5ex]
563: {\includegraphics[height=7ex]{traceTube.eps}}
564: \right._{p}
565: = \sum_{p} \sum_{r=1}^\infty
566: { \period{p} \,
567: \prpgtr{t-r \period{p}}
568: \over \oneMinJ{r} }
569: \e^{r \beta \Obser_p}
570: % \e^{r \beta \Obser_p} { \period{p} \over \oneMinJ{r} }
571: % \prpgtr{t-r \period{p}}
572: \label{tr-L1}
573: \eeq
574: which relates the spectrum of the {\evOper}
575: to a sum over prime periodic solutions $p$ of the
576: dynamical system and their repeats $r$.
577: % The trace formula \refeq{tr-L1} has an intuitive geometrical interpretation.
578:
579: What does this formula mean?
580: Prime cycles partition the dynamical space into neighborhoods,
581: each cycle enclosed by a
582: tube whose volume is the product of
583: its length $\period{p}$ and its thickness
584: $\left|\det({\bf 1}-{\bf J}_p)\right|^{-1}$.
585: The trace picks up a periodic orbit contribution only when the
586: time $t$ equals a prime period or its repeat,
587: a constraint enforced here by
588: $\prpgtr{t- r\period{p}}$.
589: ${\bf J}_p$ is the linear stability of cycle $p$,
590: so for long cycles $\oneMinJ{r} \approx$~(product of expanding eigenvalues),
591: and the contribution of long and very unstable cycles are exponentially
592: small compared to the short cycles which dominate trace formulas.
593: The number of contracting directions and the overall dimension
594: of the dynamical space is immaterial; that is why the theory
595: can also be applied to PDEs.
596: All this information is purely geometric, intrinsic to the flow,
597: coordinate reparametrization invariant,
598: and the same for any average one might wish to compute.
599: The information related to a specific observable is carried by the
600: weight
601: $\e^{\beta \Obser_p }$, the periodic orbit estimate of the contribution of
602: $\e^{\beta \Obser^t(x)}$ from the $p$-cycle neighborhood.
603:
604: The intuitive meaning of
605: %\refeq{tr-L1}:
606: a trace formula is that it expresses
607: the average $\left< \e^{\beta \Obser^t}\right>$ as a
608: discretized integral
609: \vspace{2ex}
610: \\
611: \centerline{
612: ${
613: {\mbox{smooth} \atop \mbox{dynamics}}
614: %\vspace{-6ex}
615: ~~~~~~~~
616: \hspace{-4ex}
617: \includegraphics[width=0.45\textwidth]{f_1_08_1.ps}
618: \hspace{-4ex}
619: ~~~~~~~~
620: {\mbox{linearized} \atop \mbox{neighborhoods}}
621: }$
622: }
623: \vspace{1ex}
624: \\
625: over the dynamical space partitioned topologically
626: into a repertoire of spatio-temporal patterns,
627: each weighted by the likelihood of pattern's
628: occurrence in the long time evolution of the system.
629:
630: Periodic solutions are important because they form the {skeleton}
631: of the invariant set of the long time dynamics, %\cite{cycprl,AACI},
632: with cycles ordered {hierarchically}; short cycles give good
633: approximations to the invariant set, longer cycles refinements.
634: Errors due to neglecting long cycles can be bounded, and for nice hyperbolic
635: systems they fall off exponentially or even super-exponentially
636: with the cutoff cycle length\rf{Rugh92}.
637: Short cycles can be accurately determined and global averages (such as
638: Lyapunov exponents and escape rates)
639: % and other ``thermodynamic" averages)
640: can be computed from short
641: cycles by means of {cycle expansions}.
642:
643: The Kuramoto-Sivashinsky periodic orbit calculations of
644: Lyapunov exponents and escape rates\rf{CCP96} demonstrate
645: that the periodic orbit theory
646: predicts observable averages for deterministic but classically
647: chaotic spatio-temporal systems.
648: The main problem today is not how to compute such averages ---
649: periodic orbit theory as well as direct numerical simulations
650: can handle that --- but rather that there is no consensus on {\em what}
651: the sensible experimental observables worth are predicting.
652:
653: It should be obvious, and it still needs to be said:
654: the spatio-temporally periodic
655: solutions are {\em not} to be thought of as eigenmodes,
656: a good linear
657: basis for expressing solutions of the equations of motion.
658: Something like a dilute instant approximation makes no
659: sense at all for strongly nonlinear systems that we are considering here.
660: % \[
661: % \obser(x,t) \neq \sum_p \obser_p u_p(x,t)
662: % \,.
663: % \]
664: As the equations are nonlinear,
665: the periodic solutions are in no sense additive,
666: and their linear superpositions are not solutions.
667: \vspace{2ex}
668: \\
669: \centerline{
670: $
671: {
672: A
673: \raisebox{-4.0ex}[5.5ex][4.5ex]
674: {
675: \includegraphics[height=10ex]{solution1.eps}
676: }
677: ~+~ B
678: \raisebox{-4.0ex}[5.5ex][4.5ex]
679: {
680: \includegraphics[height=10ex]{solution2.eps}
681: }
682: + \,\, \cdots \quad \neq \mbox{~~~{\em u(x,t)}}
683: \atop
684: \mbox{~~~~~~~~~~~a solution ~~~~~~~~~~~~~~ a solution
685: ~~~~~~~~~~~ not a solution}
686: }
687: $
688: }
689: \vspace{2ex}
690: \\
691: Instead,
692: it is the trace formulas and {\fd s} of the periodic
693: orbit theory that prescribe how
694: the repertoire of admissible spatio-temporal patterns is
695: to be systematically explored,
696: and how these solutions are to be
697: put together in order to predict measurable observables.
698:
699: Suppose that the above program is successfully carried out for
700: classical solutions of some field theory.
701: What are we to make of this
702: information if we are interested in the quantum behavior of
703: the system? In the semiclassical quantization
704: the classical solutions are the starting approximation.
705: % We now describe three methods for computing
706: % perturbative corrections to the leading semiclassical behavior.
707:
708: \section{Stochastic evolution}
709: \label{s-PerCorr}
710:
711: For the same pragmatic reasons that we found it profitable
712: to shy away from facing the 4-dimensional QCD head on in the
713: above exploratory foray into a strongly nonlinear field
714: theory, we shall start out by trying to understand the structure
715: of perturbative corrections for systems radically simpler than
716: a full-fledged quantum field theory.
717: %
718: First, instead of perturbative corrections to the quantum
719: problem, we shall start by exploring the perturbative corrections to
720: weakly stochastic flows.
721: %
722: Second, instead of continuous time flows, we
723: shall start by a study of a discrete time process.
724:
725: For discrete time dynamics a
726: Langevin trajectory in presence of additive noise is generated
727: by iteration
728: \beq
729: x_{n+1}=f(x_{n})+\sigma\xi_{n}
730: \,,
731: \ee{mapf(x)-Diag}
732: where $f(x)$ is a map,
733: $\xi_n$ a random variable,
734: % with the normalized distribution $P(\xi)$,
735: and $\sigma$ parametrizes the noise strength.
736: In what follows we assume that $\xi_n$ are uncorrelated,
737: and that the mapping $f(x)$ is
738: one-dimensional and expanding, but we expect that the form of
739: the results will remain the same for
740: higher dimensions, including the field theory example
741: of the preceding section.
742:
743: Tracking an individual noisy trajectory does not make much sense; what
744: makes sense is the Fokker-Planck formulation, where one
745: considers instead evolution of an ensemble of trajectories.
746: An initial density of trajectories $\phi_0(x)$ evolves with time as
747: \beq
748: \phi_{n+1}(y) =
749: \left(
750: \Lnoise{}
751: \circ
752: \phi_{n}\right)(y)
753: = \int dx \, \Lnoise{}(y,x) \phi_{n}(x)
754: \ee{DensEvol}
755: where $\Lnoise{}$ is the {\evOper}
756: \beq
757: \Lnoise{}(y,x)
758: % = \delta_\sigma(y-f(x)) \,,\quad %\continue \delta_\sigma(x) ~~
759: =
760: \int \delta(y-f(x)-\sigma \xi) P(\xi) d\xi
761: \,=\, \sigma^{-1} P\left[ \sigma^{-1}(y-f(x))\right]
762: \,,
763: \label{oper-Diag}
764: \eeq
765: and $\xi_n$ a random variable with the normalized
766: distribution $P(\xi)$, centered on $\xi = 0$.
767:
768: If the noise is weak, the goal of the theory is to compute
769: the perturbative corrections to the eigenvalues
770: $\eigenvL$ of $\Lnoise{}$
771: order by order in the noise strength $\sigma$,
772: \[
773: \eigenvL(\sigma)
774: = \sum_{m=0}^\infty \eigenvL^{(m)} {\sigma^m \over m!}
775: % \,,\qquad
776: % \eigenvL_m = {1 \over m!} \left.{d^m~ \over d\sigma^m}\eigenvL(\sigma) \right|_{\sigma=0}
777: \,.
778: \] %\ee{pertExp1}
779: One way to get at the spectrum of $\Lnoise{}$
780: is to consider the discrete
781: Laplace transform of $\Lnoise{}^n$, or the resolvent
782: \beq
783: \sum_{n=1}^\infty z^n\tr\Lnoise{}^n
784: =
785: \tr{z\Lnoise{} \over 1 - z\Lnoise{}}
786: =
787: \sum_{\alpha=0}^\infty {z\eigenvL_\alpha \over 1 - z\eigenvL_\alpha}
788: \ee{discResolv}
789: which has a pole at every $z= \eigenvL_\alpha^{-1}$.
790:
791: The effects of weak noise are of interest in their own right, as
792: any deterministic evolution that occurs in nature is affected by noise.
793: However, what is most important in the present
794: context is the fact that the form of perturbative corrections
795: for the stochastic problem is the same as for the quantum problem,
796: and still the actual calculations are
797: sufficiently simple that one can explore
798: many more orders in perturbation theory than would be possible
799: for a full-fledged field theory, and develop new perturbative
800: methods.
801:
802: The first method we try is the standard
803: Feynman-diagrammatic expansion. For semiclassical quantum mechanics of
804: a classically chaotic system such calculation was first
805: carried out by Gaspard\rf{alonso1}. The stochastic version described
806: here, implemented by Dettmann\rf{noisy_Fred},
807: reveals features not so readily apparent in
808: the quantum calculation.
809:
810: The Feynman diagram method becomes unwieldy at higher orders.
811: The second method, introduced by Vattay\rf{diag_Fred},
812: is based on Rugh's\rf{Rugh92}
813: explicit matrix representation of the {\evOper}.
814: If one is interested in evaluating numerically many orders of perturbation
815: theory and many eigenvalues,
816: this method is unsurpassed.
817:
818: The third approach, the smooth conjugations introduced
819: by Mainieri\rf{conjug_Fred}, is
820: %the most elegant one, and
821: perhaps an altogether new idea in field theory.
822: In this approach
823: the neighborhood of each saddlepoint is rectified by an appropriate
824: nonlinear field transformation, with the focus shifted
825: from the dynamics in the original field variables
826: to the properties of the conjugacy transformation.
827: The expressions obtained are equivalent to
828: sums of Feynman diagrams, but are more compact.
829:
830:
831: \section{Feynman diagrammatic expansions}
832: \label{s-FeynmDiag}
833: %{Path integrals and
834:
835: We start our computation of the weak noise corrections to the
836: spectrum of $\Lnoise{}$ by calculating the trace of the $n$-th iterate of
837: the stochastic {\evOper} $\Lnoise{}$.
838: A convenient choice of noise is Gaussian,
839: $
840: P(\xi)=
841: e^{-\xi^2/2}/{\sqrt{2\pi}}
842: \,,
843: $
844: with the trace given by an $n$-dimensional
845: integral on $n$ points along a discrete periodic chain
846: \bea
847: \tr{\Lnoise{}^n}
848: &=& \int dx_0 \cdots dx_{n-1}
849: \Lnoise{}(x_0,x_{n-1}) \cdots \Lnoise{}(x_1,x_0)
850: \continue
851: &=& \int[dx]\, \exp\left\{-\frac{1}{2\sigma^2}
852: \sum_{a}\left[x_{a+1}-f(x_a)\right]^2\right\}
853: \continue
854: &&
855: x_n = x_0 \,,\qquad [dx]=\prod_{a=0}^{n-1}{dx_a \over \sqrt{2\pi\sigma^2}}
856: \,.
857: %\nnu %\label{IntDef}
858: \label{eLnoisMtrx}
859: \eea
860: The choice of Gaussian noise is not essential,
861: as the methods that we develop here apply
862: equally well to other noise distributions, and more generally to
863: the space dependent noise distributions $P(x,\xi)$.
864: As the neighborhood of any trajectory is nonlinearly
865: distorted by the flow, the integrated noise is
866: anyway never Gaussian, but colored.
867:
868: If the classical dynamics is hyperbolic,
869: periodic solutions of given finite
870: period $n$ are isolated.
871: Furthermore, if the noise broadening $\sigma$ is
872: sufficiently weak they remain distinct, and the dominant
873: contributions come from neighborhoods of periodic points,
874: the tubes sketched in the trace formula \refeq{tr-L1}.
875: In the
876: {\em saddlepoint approximation} the trace \refeq{eLnoisMtrx} is given by
877: the sum over neighborhoods of periodic points
878: \beq
879: \tr{\Lnoise{}^{n}} \longrightarrow
880: \left. \tr{\Lnoise{}^{n}} \right|_{\mbox{\tiny sc}}
881: =
882: \sum_{x_c\inFix{n}} e^{W_c}
883: = \sum_p \cl{p} \sum_{r=1}^\infty
884: \delta_{n, \cl{p} r} e^{W_{p^r}}
885: \,.
886: \ee{SptSum1}
887: As traces are cyclic, $e^{W_c}$ is the same
888: for all periodic points in a given cycle, independent of the choice
889: of the starting point $x_c$,
890: and the periodic point sum can be rewritten in terms of
891: prime cycles $p$ and their repeats.
892: In the deterministic, $\sigma \to 0$ limit this is the discrete
893: time version of the classical trace formula \refeq{tr-L1}.
894: Effects such as noise induced tunnelling are not included in the weak
895: noise approximation.
896:
897: We now turn to the evaluation of $W_{p^r}$, the weight of
898: the $r$-th repeat of prime cycle $p$.
899: The contribution of the cycle point $x_a$ neighborhood is best
900: expressed in an intrinsic coordinate system, by
901: centering the coordinate system on the cycle points,
902: \beq
903: x_a \to x_a + \field_a
904: \,.
905: \ee{center-a}
906: From now on $x_a$ will refer to the position of the $a$-th periodic
907: point, $\field_a$ to the deviation of the noisy trajectory from
908: the deterministic one, $f_a(\field_a)$
909: to the map \refeq{mapf(x)-Diag} centered on the $a$-th cycle point,
910: and $f_a^{(m)}$ to its $m$-th derivative evaluated at the $a$-th cycle point:
911: \beq
912: f_a(\field_a) \,= \,f(x_a+\field_a) - x_{a+1}
913: \,, \qquad
914: % \continue
915: f_a'
916: =
917: f'(x_a)
918: ,\quad
919: f_a'' =
920: f''(x_a)
921: ,\quad\cdots
922: \,.
923: %\label{center-b}
924: \eeq
925: Rewriting the trace in vector notation, with
926: $x$ and $f(x)$ $n$-dimensional
927: column vectors with components $x_a$ and $f(x_a)$
928: respectively,
929: expanding $f$ in Taylor series around each of the periodic points
930: in the orbit of $x_c$, separating out the quadratic part
931: and integrating we obtain
932: \bea
933: e^{W_c} &=& \int_c[d\field]\,
934: e^{-\left(\InvPrpgtr{}\field_{} - V'(\field) \right)^2/2\sigma^2}
935: = \int_c[d\field]\,
936: e^{-{1 \over 2\sigma^2} \,
937: \field_{}^T{1 \over {\Prpgtr{}}^T\Prpgtr{}}\field_{}
938: \,+\, (\cdots)
939: }
940: \continue
941: &=& |\det \Prpgtr{}| \int_c[d\varphi]\,
942: e^{\sum{1\over k} \tr\left(\Prpgtr{}V''(\field) \right)^k}
943: e^{-\varphi^2/2\sigma^2}
944: \label{eWcMtrx}
945: \eea
946: The [$n$$\times$$n$] matrix $\Prpgtr{}$ arises from the quadratic
947: part of the exponent, while all higher powers of $\field_a$
948: are collected in $V(\field)$:
949: \beq
950: \InvPrpgtr{ab}\field_{b} = -\Df{a}\field_{a}+\field_{a+1}
951: \,,\qquad
952: V(\field)= \sum_a \sum_{m=2}^{\infty}f^{(m)}_{a}
953: \frac{\field_{a}^{m+1}}{(m+1)!}
954: \,.
955: \ee{DefPrpg}
956: The saddlepoint expansion is most conveniently evaluated in terms of
957: Feynman diagrams, by
958: drawing $\Prpgtr{}$ as a directed line
959: $ \Prpgtr{ab} = \btrack{Delta}$,
960: and the derivatives of $V$ as the ``interaction'' vertices
961: \PC{fix birdtrack height}
962: \[
963: f^{''}_a
964: \,=\,
965: \btrack{f2}
966: \,,\quad
967: f^{'''}_a
968: \,=\,
969: \btrack{f3}
970: \,,\quad
971: \dots
972: \,.
973: \]
974: %\vspace{0.5\Fsize}
975: In the jargon of field theory,
976: $\Prpgtr{}$ is the ``free propagator''.
977: Its determinant
978: \beq
979: \left|\det\,\Prpgtr{}\right|={1 \over |\ExpaEig_c-1|}
980: \,, \qquad
981: \ExpaEig_c = \prod_{a=0}^{n-1}\Df{a}
982: \ee{detDel}
983: is the 1-dimensional version of the classical stability weight
984: $\left|\det({\bf 1}-{\bf J})\right|^{-1}$
985: in \refeq{tr-L1},
986: with $\ExpaEig_c$ the {stability} of the $n$-cycle going through
987: the periodic point $x_c$.
988:
989: Standard methods\rf{FieldThe}
990: now yield the perturbation expansion
991: in terms of the connected ``vacuum bubbles''
992: \bea
993: W_c &=& - \ln|\ExpaEig_c-1| + \sum_{k=1}^\infty W_{c,2k}\sigma^{2k}
994: \label{e:PertExpW}\\
995: W_{c,2} &=&
996: % {1\over 2} %\btrack{dumbell1}
997: % + {1\over 2}% \btrack{dumbell2}
998: % + {1\over 2}% \btrack{infty}
999: % + {1\over 2}% \btrack{eye}
1000: %\,.
1001: %\left\{
1002: {1 \over 2}\,
1003: \btrack{FeynmDumb1} \,+\,
1004: {1 \over 2}\,
1005: \btrack{FeynmDumb2} \,+\,
1006: {1 \over 2}\,
1007: \btrack{FeynmEight} \,+\,
1008: {1 \over 2}\,
1009: \btrack{FeynmLips}
1010: %\right\}
1011: \label{a:sigSqGraphs}
1012: \continue
1013: W_{c,4} &=& \cdots
1014: \nnu
1015: \,.
1016: \eea
1017: In the usual field-theoretic calculations the $ W_{c,0}$ term
1018: corresponds to an overall volume term that cancels out in the expectation
1019: values. In contrast, as explained in \refsect{PerOrsInAction}, here the
1020: $ e^{W_{c,0}} = |\ExpaEig_c-1|^{-1}$ term
1021: is the classical volume of cycle $c$.
1022: Not only does this weight not
1023: cancel out in the expectation value formulas, it
1024: plays the key role both in classical and
1025: semiclassical trace formulas.
1026:
1027: In the diagrams sketched above
1028: a propagator line connects $x_a$ at time $a$ with
1029: $x_{b}$ at later time $b$ by a deterministic trajectory. At time
1030: $b$ noise induces a kick whose strength depends on the local
1031: curvature of the flow. A penalty of a factor $\sigma$ is paid,
1032: $m-1$ deterministic trajectories originate in
1033: the neighborhood of $x_{b}$ from vertex $V^{(m)}(x_{b})$,
1034: and the process repeats itself, each vertex carrying a penalty
1035: of $\sigma$, and higher derivatives of the $f_b$.
1036: Summing over all noise kick sequences encoded by
1037: a given diagram and using the periodicity of the trace integral
1038: \refeq{eLnoisMtrx} Dettmann\rf{noisy_Fred} obtains expressions such as
1039: \beq
1040: \frac{r}{2}\frac{\ExpaEig_p^{2r}-1}{\ExpaEig_p^2-1}
1041: \frac{\ExpaEig_p^r}{(\ExpaEig_p^r-1)^3}
1042: \sum_{\mod{a}\mod{b}}
1043: \left(
1044: \frac{f^{''2}_{\mod{a}}}{f^{'2}_{\mod{a}}}
1045: -\frac{f^{'''}_{\mod{a}}}{\Df{\mod{a}}}
1046: \right)
1047: \prod_{\mod{d}=\mod{b}+1}^{\mod{a}-1}f^{'2}_{\mod{d}}
1048: \,.
1049: \ee{FynmEight}
1050: This particular sum is the $\btrack{FeynmEight}$
1051: Feynman diagram $\sigma^2$ correction to $r$-th repeat of prime cycle
1052: $p$.
1053: More algebra leads to similar contributions from
1054: the remaining diagrams. But the overall result is surprising;
1055: the dependence on the repeat number $r$ factorizes, with
1056: each diagram yielding the same prefactor
1057: depending only on $\ExpaEig_p^r$.
1058: This remarkable fact will be explained in \refsect{scfpo}.
1059: The result of the Feynman-diagrammatic
1060: calculations is the {\em stochastic trace formula}
1061: \beq
1062: \left.
1063: \tr{z\Lnoise{} \over 1 - z\Lnoise{}}
1064: \right|_{\mbox{\tiny sc}}
1065: =
1066: \sum_p\sum_{k=0}^{\infty}
1067: {\cl{p} \, t_{p,k} \over 1-t_{p,k}}
1068: \,,\quad
1069: t_{p,k}
1070: =
1071: \frac{z^\cl{p}}{|\ExpaEig_p|\ExpaEig_p^k}
1072: e^{\frac{\sigma^2}{2} % (k+1)^2 w_{p,2}
1073: w_{p,k}^{(2)}
1074: + O(\sigma^4)
1075: }
1076: \,,
1077: \label{NoiseResum}
1078: \eeq
1079: where $t_{p,k}$ is the $k$-th local eigenvalue evaluated on the $p$ cycle.
1080: The deterministic, $\sigma=0$ part of this formula
1081: is the stochastic equivalent of the Gutzwiller semiclassical
1082: trace formula\rf{gutbook}. The $\sigma^2$ correction $w_{p,k}^{(2)}$ is
1083: the stochastic analogue of Gaspard's $\hbar$ correction\rf{alonso1}.
1084: At the moment the explicit formula
1085: % for $w_{p,k}^{(2)}$
1086: is sufficiently unenlightening that we postpone writing it down to
1087: \refsect{scfpo}.
1088:
1089:
1090: While the diagrams are standard,
1091: the chaotic field theory
1092: calculations are considerably more demanding than is usually the
1093: case in field theory. Here there is
1094: no translational invariance along the chain, so
1095: the vertex strength depends on the position, and
1096: the free propagator is not diagonalized by a Fourier transform.
1097: %The computations carried out so far were performed
1098: %in configuration space.
1099: %Unlike the most field-theoretic literature,
1100: Furthermore,
1101: here one is neither ``quantizing'' around a trivial vacuum,
1102: nor a countable infinity of analytically explicit
1103: soliton saddles, but around an
1104: infinity of nontrivial unstable hyperbolic saddles.
1105:
1106: Two aspects of the above
1107: perturbative results are {\em a priori} far from obvious:
1108: (a) that the structure of the periodic orbit theory
1109: should survive introduction of noise, and (b)
1110: a more subtle and surprising result,
1111: repeats of prime cycles can be re-summed and theory reduced to the
1112: \dzeta s and \fd s of the same form as for deterministic
1113: systems.
1114:
1115: Pushing the Feynman-diagrammatic approach to higher orders is laborious,
1116: and has not been attempted for this class of problems.
1117: As we shall now see, it is not smart to keep pushing it, either,
1118: as one can compute many more orders of perturbation theory
1119: by means of a matrix representation for $\Lnoise{}$.
1120:
1121:
1122: \section{{\EvOper} in a matrix representation}
1123: \label{s-MatrixRep}
1124:
1125: An expanding map $f(x)$ takes an initial smooth
1126: distribution $\phi(x)$ defined on a subinterval,
1127: stretches it out and overlays it over a larger interval.
1128: Repetition of this process smoothes the initial
1129: distribution $\phi(x)$, so it is natural to concentrate on
1130: smooth distributions $\phi_{n}(x)$, and represent them
1131: by their Taylor series.
1132: By expanding both $\phi_{n}(x)$ and $\phi_{n+1}(y)$ in
1133: \refeq{DensEvol} in Taylor series Rugh\rf{Rugh92} derived a matrix
1134: %$\Lmat{}$
1135: representation of the {\evOper}
1136: \[
1137: \int dx \, \Lnoise{}(y,x) \frac{x^{m}}{m!}
1138: = \sum_{m'} \frac{y^{m'}}{m'!} \Lmat{}_{m'm}
1139: \,,\qquad m,m'= 0,1,2, \dots
1140: \]
1141: which maps the $x^m$ component of the density of trajectories $\phi_n(x)$
1142: in \refeq{DensEvol} to the $y^{m'}$ component of
1143: the density $\phi_{n+1}(y)$ one time step later.
1144: % with $y=f(x)$.
1145: % An analytic function $g(x)$ has a Taylor series expansion
1146: % \[
1147: % g(y) = \sum_{m=0}^\infty
1148: % \frac{y^{m}}{m!} \left. \frac{\pde^{m}}{\pde x^{m}}g(x) \right|_{x=0}
1149: % \,,
1150: % \]
1151: The matrix elements follow by differentiating both sides
1152: with $\pde^{m'}/\pde y^{m'}$ and evaluating the integral
1153: \beq
1154: \Lmat{}_{m'm} =
1155: \left. \frac{\pde^{m'}}{\pde y^{m'}}
1156: \int dx \, \Lnoise{}(y,x)
1157: \frac{x^{m}}{m!} \right|_{y=0}
1158: \,.
1159: \ee{Lmat-Diag}
1160:
1161:
1162: % \subsection{Weak noise expansion of the {\evOper}}
1163:
1164: % \subsection{Local matrix representation of {\evOper}}
1165: % \label{LocMatr-Diag}
1166:
1167: In \refeq{oper-Diag} we have written the {\evOper} $\Lnoise{}$ in terms of
1168: the Dirac delta function
1169: % ,
1170: % $
1171: % \Lnoise{}(x',x)=\int \delta(x'-f(x)-\sigma\xi)p(\xi)d\xi
1172: % $,
1173: in order to emphasize that
1174: in the weak noise limit the stochastic trajectories are concentrated
1175: along the classical trajectory $y = f(x)$.
1176: Hence it is natural to
1177: expand the kernel in a Taylor series~\rf{Watanabe87}
1178: in $\sigma$
1179: \beq
1180: \Lnoise{}(y,x) =
1181: \delta(y-f(x))
1182: +\,
1183: \sum_{n=2}^{\infty}\frac{(-\sigma)^n}{n!}\delta^{(n)}(y-f(x))
1184: \int \xi^n P(\xi)d\xi
1185: \,,
1186: \ee{opexp}
1187: where
1188: $
1189: \delta^{(n)}(y) = {\pde^n \over \pde y^n} \delta(y)
1190: \,.
1191: $
1192: This yields a representation of the
1193: {\evOper} centered along the classical trajectory,
1194: dominated by the deterministic
1195: {\FPoper} $\delta(y-f(x))$, with corrections given by derivatives
1196: of delta functions weighted by moments of the noise
1197: distribution $P_n=\int P(\xi)\xi^nd\xi$.
1198: % \beq
1199: % \Lnoise{}(y,x)
1200: % =
1201: % \delta(y-f(x)) \,+\,
1202: % \sum_{n=2}^{\infty}\frac{(-\sigma)^n}{n!} P_n \delta^{(n)}(y-f(x)).
1203: % \ee{opexp1}
1204: We again center the coordinate system on the cycle points
1205: as in \refeq{center-a},
1206: and also introduce a notation for the operator \refeq{oper-Diag}
1207: centered on the $x_a \to x_{a+1}$ segment of the
1208: classical trajectory
1209: \[
1210: \Lnoise{a}(\field_{a+1},\field_a)
1211: \,=\,
1212: \Lnoise{}(x_{a+1}+\field_{a+1},x_a+\field_a)
1213: \,.
1214: \]
1215: The weak noise expansion \refeq{opexp} for the $a$-th segment operator
1216: is given by
1217: \beq
1218: \Lnoise{a}(\field_{a+1},\field_a)=
1219: \delta(\field_{a+1}- f_a(\field_a))
1220: +\sum_{n=2}^{\infty}\frac{(- \sigma)^{n}}{n!}
1221: P_n \delta^{(n)}(\field_{a+1}- f_a(\field_a))
1222: \,.
1223: \label{LnoiseExp}
1224: \eeq
1225: As the {\evOper} has a simple $\delta$-function form,
1226: the local matrix representation of $\Lnoise{a}$ centered on
1227: the $x_a \to x_{a+1}$ segment of the deterministic trajectory
1228: can be evaluated recursively in terms of derivatives of the map $f$:
1229: \bea
1230: \left(\Lmat{a}\right)_{m'm}
1231: &=&
1232: \sum_{n}^{\infty}
1233: P_n\frac{(-\sigma)^n}{n!}(\Bmat{a})_{m'+n,m}
1234: \,,\qquad n= \mbox{max}(m-m',0)
1235: %\label{BtoL}
1236: \continue
1237: (\Bmat{a})_{m' m}
1238: % &=&
1239: % \left. \frac{\pde^{n}}{\pde \field'^{n}}
1240: % \int d\field \,
1241: % \delta(\field' -f_a(\field))
1242: % \frac{\field^{m}}{m!}
1243: % \right|_{\field'=0}
1244: % \continue
1245: &=&
1246: \int d\field \,
1247: \delta^{(m')}(\field_{a+1} -f_a(\field)) \frac{\field^{m}}{m!}
1248: \label{Bmatrix}\\
1249: %\continue
1250: &=&
1251: \left.
1252: {1 \over |f_a'|}
1253: \left( {d~ \over d\field} {1 \over f_a'(\field)}\right)^{m'}
1254: \frac{\field^{m}}{m!}
1255: \right|_{\field=0}
1256: \,.
1257: \nnu
1258: \eea
1259: The matrix elements vanish for $m'<m$, so $\Bmat{}$ is
1260: a lower triangular matrix.
1261: The diagonal and the successive off-diagonal matrix elements
1262: are easily evaluated iteratively by computer algebra
1263: \[
1264: (\Bmat{a})_{m m} = \frac{1}{|f_a'|(f_a')^{m}}
1265: \,,\quad
1266: (\Bmat{a})_{m+1,m} = - \frac{(m+2)! f_a''}{ 2 m! |f_a'|(f_a')^{m+2}}
1267: \label{Bexplicit}
1268: \,,~~\cdots
1269: \,.
1270: \]
1271: For chaotic systems the map is expanding, $|f_a'|>1$. Hence
1272: the diagonal terms drop off exponentially, as $1/|f_a'|^{m+1}$,
1273: the terms below the diagonal fall off even faster, and
1274: truncating $\Lmat{a}$ to a finite matrix
1275: introduces only exponentially small errors.
1276:
1277: The trace formula \refeq{discResolv}
1278: % \refeq{eLnoisMtrx}
1279: takes now a matrix form
1280: %$ \Lmat{}$
1281: \beq
1282: \left.
1283: \tr{z\Lnoise{} \over 1 - z\Lnoise{}}
1284: \right|_{\mbox{\tiny sc}}
1285: =
1286: \sum_p \cl{p} \tr {z^\cl{p} \Lmat{p} \over 1-z^\cl{p} \Lmat{p} }
1287: \,,
1288: \label{tracenp}
1289: \eeq
1290: where
1291: $ \Lmat{p} = { \Lmat{\cl{p}}\Lmat{2}\cdots \Lmat{1}}$
1292: is the contribution of the $p$ cycle.
1293: The subscript {\tiny sc} is a reminder that this is a saddlepoint
1294: or semiclassical approximation,
1295: % to $\tr {\Lnoise{}}^n$,
1296: valid as an asymptotic series in the limit of weak noise.
1297: Vattay\rf{VatBS}
1298: interprets the local matrix representation of the \evOper\ as follows.
1299: The matrix identity log~det~=~tr~log together with
1300: the trace formula \refeq{tracenp} yields
1301: \bea
1302: \left.
1303: \det(1 - z \Lop)
1304: \right|_{\mbox{\tiny sc}}
1305: =
1306: \prod_{p} \det(1 - z^{\cl{p}} \Lmat{p})
1307: \,,
1308: \label{fredholmprod}
1309: \eea
1310: so in
1311: the saddlepoint approximation the spectrum of the {\em global}
1312: {\evOper} $\Lnoise{}$ is
1313: pieced together from the {\em local} spectra computed cycle-by-cycle
1314: on neighborhoods of individual prime cycles with periodic boundary
1315: conditions. The meaning of the $k$-th term in
1316: the trace formula \refeq{NoiseResum} is now clear; it is the
1317: $k$-th eigenvalue of the local {\evOper} restricted to the
1318: $p$-th cycle neighborhood.
1319:
1320: Using this matrix representation Palla and S\o ndergaard\rf{diag_Fred}
1321: were able to compute corrections to order $\sigma^{12}$,
1322: a feat simply impossible along the Feynman-diagrammatic
1323: line of attack.
1324: In retrospect, the matrix representation method for solving the
1325: stochastic evolution is eminently sensible --- after all, that is
1326: the way one solves a close relative to stochastic PDEs,
1327: the Schr\"odinger equation. What is
1328: new is that the problem is being solved locally,
1329: periodic orbit by periodic orbit, by translation to
1330: coordinates intrinsic to the periodic orbit. It
1331: is this natural local basis that makes the
1332: matrix representation so simple.
1333:
1334: Mainieri\rf{conjug_Fred}
1335: takes this observation one step further; as the dynamics is
1336: nonlinear, why not search for a nonlinear coordinate transformation that
1337: makes the intrinsic coordinates as simple as possible?
1338:
1339: \section{Smooth conjugacies}
1340: \label{scfpo}
1341:
1342: This step
1343: % might be unfamiliar to a field theorist, but is
1344: injects into field theory a method
1345: standard in the construction of normal forms for bifurcations\rf{KatHass}.
1346: The idea is to perform a smooth nonlinear coordinate transformation
1347: $x = h(y)$,
1348: $ f(x) = h(g(h^{-1}(x)))$
1349: that flattens out the vicinity
1350: of a fixed point and makes the map {\em linear} in an open neighborhood,
1351: $ f(x) \to g(y) = {\bf J} \cdot y$.
1352: \vspace{2ex}
1353: \\
1354: \centerline{
1355: ${
1356: \raisebox{-4.0ex}[5.5ex][4.5ex]
1357: {
1358: \includegraphics[height=10ex]{conjug-a.eps}
1359: }
1360: \atop
1361: \mbox{an arbitrary coordinatization}
1362: }$
1363: ~~~
1364: $\Longrightarrow$
1365: ~~~
1366: ${
1367: \raisebox{-4.0ex}[5.5ex][4.5ex]
1368: {
1369: \includegraphics[height=10ex]{conjug-b.eps}
1370: }
1371: \atop
1372: \mbox{intrinsic, flat coordinates}
1373: }$
1374: }
1375: \vspace{2ex}
1376: \\
1377: The key idea of flattening the neighborhood of
1378: a saddlepoint can be traced back to
1379: Poincar\'e's celestial mechanics,
1380: and is perhaps not something that a field theorist would instinctively
1381: hark to as a method of computing perturbative corrections.
1382: % \PC{mention period doubling operator spectrum}
1383: This local rectification of a map can be implemented
1384: only for isolated non-degenerate fixed points (otherwise
1385: higher terms are required by the normal form expansion around the
1386: point), and only in finite neighborhoods, as the
1387: conjugating functions in general have finite radia of convergence.
1388: %\PC{remark that $h^{-1}$ is entire if $f$ is?}
1389:
1390: We proceed in two steps. First, substitution of
1391: the weak noise perturbative
1392: expansion of the {\evOper} \refeq{LnoiseExp} into the trace
1393: centered on cycle $c$ generates products of derivatives of
1394: $\delta$-functions:
1395: \[
1396: \left. \tr{\Lnoise{}^{n}} \right|_{c}
1397: = \cdots + \int[d\field]\,
1398: \left\{\cdots
1399: \delta^{(m')}(\field''- f_a(\field'))
1400: \,
1401: \delta^{(m)}(\field'- f_{a-1}(\field))
1402: \cdots \right\}
1403: + \cdots
1404: \,.
1405: %\label{LnsExp}
1406: \]
1407: The integrals are evaluated as in \refeq{Bmatrix}, yielding
1408: recursive derivative formulas such as
1409: \beq
1410: \int dx \,
1411: \delta^{(m)}(y)
1412: =
1413: \left.
1414: {1 \over |y'(x)|}
1415: \left(- {d~ \over dx} {1 \over y'(x)}
1416: \right)^{m}
1417: \right|_{y=0}
1418: \,,\qquad
1419: y=f(x)-x
1420: \,.
1421: \ee{recurDer}
1422: or $n$-point integrals, with derivatives distributed
1423: over $n$ different $\delta$-functions.
1424:
1425: Next we linearize the neighborhood of the $a$-th cycle point.
1426: For a 1-dimensional map $f(x)$ with a fixed point $f(0)=0$
1427: of stability $\ExpaEig = f'(0)$, $|\ExpaEig|\neq 1$ we search
1428: for a smooth conjugation $h(x)$
1429: such that:
1430: \beq
1431: f(x) = h(\ExpaEig h^{-1}(x))
1432: \,,\quad
1433: h(0)=0
1434: \,,\quad
1435: h'(0)=1
1436: \,.
1437: \label{EqConj}
1438: \eeq
1439: In higher dimensions $\ExpaEig$ is replaced by the Jacobian matrix
1440: ${\bf J}$.
1441: % and conjugation requires that the stability eigenvalues are non-resonant.
1442: For a periodic orbit each point around the cycle has a
1443: differently distorted neighborhood,
1444: with differing second and higher derivatives,
1445: so the conjugation function $h_a$ has to be computed
1446: point by point,
1447: \[
1448: f_a(\phi)=h_{a+1}(f'_a h_a^{-1}(\phi))
1449: \,.
1450: \]
1451: An explicit expression for $h_a$ in terms of $f$ is
1452: obtained by iterating around the whole cycle, and using the chain
1453: rule \refeq{detDel} for the cycle stability $\ExpaEig_p$
1454: %\PC{recheck this!}
1455: \beq
1456: f_a^{\cl{p}}(\phi)=h_a(\ExpaEig_p h_a^{-1}(\phi))
1457: \,,
1458: \ee{hCycle}
1459: so each $h_a$ is given by some combination
1460: of $f_a$ derivatives along the cycle.
1461: Expand $f(x)$ and $h(x)$
1462: \[
1463: f(x)
1464: =
1465: % \,&=&
1466: \ExpaEig x + x^2 f_2 + x^3f_3 + \dots
1467: \,,\quad
1468: % \continue
1469: h(y)
1470: =
1471: % \,&=&
1472: y + y^2 h_2 + y^3 h_3 + \dots
1473: \;\,,
1474: % \label{f-hExps}
1475: \]
1476: and equate recursively coefficients in the
1477: functional equation $ h(\ExpaEig y)=f(h(y))$ expansion
1478: \beq
1479: h(\ExpaEig u) - \ExpaEig h(u) =
1480: \sum_{n=2}^\infty f_m \left(h(u)\right)^m
1481: \;\,.
1482: \label{EqConjExp}
1483: \eeq
1484: This yields the expansion %\rf{KatHass}
1485: for the conjugation function $h$
1486: in terms of the mapping $f$
1487: \beq
1488: h_2 = \frac{f_2}{ \ExpaEig (\ExpaEig - 1) }
1489: \,,\qquad
1490: h_3 =
1491: \frac{2 f_2^2 + \ExpaEig ( \ExpaEig -1 ) f_3}
1492: { \ExpaEig^2
1493: ( \ExpaEig -1 )
1494: ( \ExpaEig^2 -1 )
1495: }
1496: \,, \qquad\cdots
1497: \;\,.
1498: \ee{h(2)}
1499: The periodic orbit
1500: conjugating functions $h_a$ are obtained in the same way
1501: from \refeq{hCycle},
1502: %by equating coefficients of Taylor series, and requiring
1503: with proviso that
1504: the cycle stability is not marginal, $|\ExpaEig_p|\neq 1$.
1505:
1506: What is gained by replacing the perturbation expansion
1507: in terms of $f^{(m)}$ by still
1508: messier perturbation expansion for the
1509: conjugacy function $h$?
1510: Once the
1511: neighborhood of a fixed point is linearized, the conjugation formula
1512: for the repeats of the map
1513: \[
1514: f^r(x)=h(\ExpaEig^r h^{-1}(x))
1515: \]
1516: can be used to compute
1517: derivatives of a function composed with itself $r$ times.
1518: The expansion for arbitrary number of repeats depends
1519: on the conjugacy function $h(x)$
1520: computed for a {\em single} repeat, and all the dependence on the
1521: repeat number is carried by polynomials in $\ExpaEig^r$,
1522: a result that emerged as a surprise in the Feynman diagrammatic
1523: approach of \refsect{s-FeynmDiag}.
1524: The integrals such as \refeq{recurDer} evaluated on the $r$-th repeat of
1525: prime cycle $p$
1526: \beq
1527: y(x) = f^{\cl{p}r}(x) - x
1528: \ee{g(x)}
1529: have a simple dependence
1530: on the conjugating function $h$
1531: {\small
1532: \bea
1533: %\left.
1534: \frac{1}{3!}\frac{\partial^2}{\partial y^2}
1535: \frac{1}{y'(0)}
1536: %\right|_{x=0}
1537: &=&
1538: {\frac{{{\ExpaEig }^r}\left( 1 + {{\ExpaEig }^r} \right)
1539: }{{{\left( \ExpaEig^r -1 \right)
1540: }^3}}}
1541: \left(2h_2^2 - h_3 \right)
1542: %\ee{der(2)}
1543: \label{der(5)} \\
1544: \continue
1545: %\left.
1546: \frac{1}{4!}\frac{\partial^3}{\partial y^3}
1547: \frac{1}{y'(0)}
1548: %\right|_{x=0}
1549: &=&
1550: -5\ExpaEig^r{ {( \ExpaEig^r + 1 )^2}
1551: \over
1552: (\ExpaEig^r - 1)^4 }h_2^3
1553: + \ExpaEig^r { 5\ExpaEig^{2r} + 8\ExpaEig^r + 5
1554: \over
1555: (\ExpaEig^r - 1)^4 } h_2h_3
1556: % \ceq
1557: %~~~~~~~~~~~~~~~~~~~~~~
1558: - \ExpaEig^r { \ExpaEig^{2r} + \ExpaEig^r + 1
1559: \over
1560: (\ExpaEig^r - 1)^4 } h_4
1561: \continue
1562: \cdots &=&\cdots
1563: %\label{der(3)}
1564: \nnu
1565: %\ee{der(3)}\\
1566: \eea
1567: }%end \small
1568: The evaluation of $n$-point integrals is more subtle\rf{conjug_Fred}.
1569: The final result of all these calculations is that
1570: % in (\ref{der(2)})
1571: expressions of form \refeq{der(5)} depend
1572: on the conjugation function determined from the
1573: iterated map,
1574: % with $h$ carrying all dependence on the
1575: % higher derivatives of the original map,
1576: % replaces the sum of Feynman diagrams,
1577: with the saddlepoint approximation to
1578: the {\fd} given by
1579: \[
1580: \left. \det(1-z{\Lop}_\sigma)\right|_{\mbox{\tiny sc}}
1581: =
1582: \prod_p\prod_{k=0}^{\infty}(1-t_{p,k})
1583: \]
1584: in terms of local $p$-cycle eigenvalues
1585: \bea
1586: t_{p,k}&=&\frac{z^{\cl{p}}}{|\ExpaEig_p|\ExpaEig_p^k}
1587: \e^{\frac{\sigma^2}{2}P_2 w^{(2)}_{p,k}
1588: +\frac{\sigma^3}{3!}P_3 w^{(3)}_{p,k}
1589: +\frac{\sigma^4}{4!}P_4 w^{(4)}_{p,k}+O(\sigma^6)}
1590: \continue
1591: w^{(2)}_{p,k}&=&(k+1)^2\sum_a (2h_{a,2}^{2} - h_{a,3})
1592: \,,\qquad
1593: w^{(3)}_{p,k} = \cdots
1594: \,,\cdots \,.
1595: \nnu
1596: \eea
1597: accurate up to order $\sigma^4$.
1598: $w^{(3)}$, $w^{(4)}$ are also computed by Dettmann,
1599: but we desist from citing them here; the reader is
1600: referred to
1601: % contribution $w^{(3)}_{p,k}$ is given in
1602: \refref{conjug_Fred}.
1603: What is remarkable about these results is their simplicity when
1604: expressed in terms of the conjugation function $h$, as opposed
1605: to the Feynman diagram sums, in which each diagram contributes
1606: a sum like the one in \refeq{FynmEight}, or worse. Furthermore, both
1607: the conjugation and the matrix approaches are easily automatized,
1608: as they require only recursive evaluation of derivatives, as
1609: opposed to the handcrafted Feynman diagrammar.
1610:
1611: Simple minded as they might seem, discrete stochastic processes are
1612: a great laboratory for testing ideas that would otherwise be hard to
1613: test. Dettmann, Palla and S\o ndergaard have
1614: used a 1-dimensional repeller of bounded nonlinearity
1615: and complete binary symbolic dynamics to check numerically
1616: the above results, and computed the leading eigenvalue of
1617: $\Lnoise{}$ by no less than five different methods.
1618: As anticipated by Rugh\rf{Rugh92},
1619: the {\evOper} eigenvalues
1620: converge super-exponentially with the cycle length; addition
1621: of cycles of period $(n$+1$)$ to the set of all cycles
1622: up to length $n$ {\em doubles} the number of significant
1623: digits in the perturbative prediction.
1624: However, as the series is asymptotic, for realistic values
1625: of the noise strength summations beyond all orders are
1626: needed\rf{asym_Fred}.
1627:
1628:
1629: \section{Summary}
1630:
1631: The periodic orbit theory approach to turbulence is to visualize
1632: turbulence as a sequence of near recurrences in a repertoire of unstable
1633: spatio-temporal patterns.
1634: The investigations of the Kuramoto-Sivashinsky system
1635: discussed above are first
1636: steps in the direction of implementing this program.
1637: So far, existence of a
1638: hierarchy of spatio-temporally periodic solutions
1639: of spatially extended nonlinear system has been demonstrated,
1640: and the periodic orbit theory has been tested in
1641: evaluation of global averages for such system.
1642: The parameter ranges tested so far probe
1643: the weakest nontrivial ``turbulence'', and it is an open
1644: question to what extent the approach remains implementable
1645: as the system goes more turbulent.
1646:
1647: The most important lesson of this investigation is that
1648: the unstable spatio-temporally periodic
1649: solutions do explore systematically the
1650: repertoire of admissible spatio-temporal patterns, with the trace
1651: and {\fd s} formulas and their cycle expansions being the proper tools for
1652: extraction of quantitative predictions from the periodic orbits data.
1653:
1654: We formulat next a semiclassical perturbation theory for
1655: stochastic trace formulas
1656: with support on infinitely many chaotic saddles.
1657: The central object of the periodic orbit theory, the trace of the {\evOper},
1658: is a discrete path integral, similar to those found in field theory
1659: and statistical mechanics. The weak noise perturbation theory,
1660: likewise, resembles perturbative field theory, and can be cast
1661: into the standard field-theoretic language of Feynman diagrams.
1662: However, we found out that both the matrix and the nonlinear
1663: conjugacy perturbative methods are superior to the standard approach.
1664: In contrast to previous perturbative expansions around
1665: vacua and instanton solutions, the location
1666: and local properties of each saddlepoint must
1667: be found numerically.
1668:
1669: The key idea in the new formulation of perturbation
1670: theory is this: Instead of separating the action into
1671: quadratic and ``interaction'' parts, one first performs a nonlinear
1672: field transformation which turns the saddle point into an exact
1673: quadratic form. The price one pays for this is the Jacobian of
1674: the nonlinear field transformation ---
1675: but it turns out
1676: that the perturbation expansion of this Jacobian
1677: {in terms of the conjugating function} is order-by-order
1678: more
1679: compact than the Feynman-diagrammatic expansion.
1680:
1681: \section*{Acknowledgements}
1682: I am indebted to my collaborators
1683: C.P.~Dettmann, G.~Vattay,
1684: F.~Christiansen,
1685: V.~Putkaradze,
1686: G.~Palla, N.~S\o ndergaard,
1687: R.~Mainieri
1688: and
1689: H.H.~Rugh
1690: for co-suffering through all the details omitted in
1691: this overview.
1692: I am grateful to
1693: E.A. Spiegel, L.~Tuckerman and M.J.~Feigenbaum
1694: for patient instruction. I am not grateful to those
1695: directors and gentlemen of committees who
1696: do not find theoretical physics a
1697: vibrant subject.
1698:
1699:
1700: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1701: %%% APPENDICES
1702: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1703:
1704: % \appendix
1705:
1706: % \section{Algebra}
1707:
1708:
1709: %REFERENCES -------------------------------------------------------
1710:
1711: \begin{thebibliography}{999}
1712:
1713:
1714: \bibitem{KinoHughes} V.W. Hughes and T. Kinoshita,
1715: % ANOMALOUS G VALUES OF THE ELECTRON AND MUON.
1716: {\em Rev. Mod. Phys. \bf 71} (1999) S133. %:S133-S139
1717:
1718: \bibitem{Dehmelt87} R.S. Van Dyck, Jr., P.B. Schwinberg, H.G. Dehmelt,
1719: {\em Phys. Rev. Lett. \bf 59} (1987) 26.
1720:
1721: \bibitem{KWilson} K.G. Wilson,
1722: % Lattice gauge theories introduced
1723: {\em Phys. Rev. \bf D 10} (1974) 2445.
1724:
1725: \bibitem{Polyakov} A.A.~Belavin, A.M. Polyakov, A.S.~Swartz, Yu.S.~Tyupkin,
1726: % SU(2) instantpons discovered
1727: {\em Phys. Lett. \bf B 59} (1975) 85.
1728:
1729: \bibitem{gutbook} M.C. Gutzwiller, {\em Chaos in Classical
1730: and Quantum Mechanics} (Springer, New York 1990).
1731:
1732: \bibitem{KurSiv}
1733: % \bibitem{Kur}
1734: Y. Kuramoto and T. Tsuzuki,
1735: % Persistent propagation of concentration waves in
1736: % dissipative media far from thermal equilibrium
1737: {\em Progr. Theor. Physics \bf 55} (1976) 365;
1738: \\
1739: % \bibitem{Siv}
1740: G.I. Sivashinsky,
1741: % {Nonlinear analysis of hydrodynamical instability
1742: % in laminar flames --- I. Derivation of basic equations}
1743: {\em Acta Astr. \bf 4} (1977) 1177.
1744:
1745: \bibitem{CCP96} F. Christiansen, P. Cvitanovi\'c, V. Putkaradze,
1746: % {\em ``Hopf's last hope:
1747: % spatio-temporal chaos in terms of unstable recurrent patterns''},
1748: {\em Nonlinearity 10} (1997) 1.
1749:
1750: \bibitem{ZG96} S.M. Zoldi and H.S. Greenside,
1751: %``Spatially localized unstable periodic orbits
1752: % of a high-dimensional chaotic system'',
1753: {\em Phys. Rev. \bf E 57} (1998) R2511. % R2511-R2514
1754:
1755: \bibitem{QCcourse} P. Cvitanovi\'c et al.,
1756: {\em Classical and Quantum Chaos}
1757: (Niels Bohr Institute, Copenhagen 1999);
1758: {\tt www.nbi.dk/ChaosBook/}.
1759:
1760: \bibitem{Rugh92} H.H. Rugh,
1761: % ``The Correlation Spectrum for Hyperbolic Analytic Maps'',
1762: {\em Nonlinearity \bf 5} (1992) 1237.
1763:
1764: \bibitem{alonso1}
1765: P.~Gaspard and D.~Alonso,
1766: % {$\hbar$} expansion for the periodic-orbit quantization of hyperbolic
1767: % systems.
1768: {\em Phys. Rev. \bf A 47} (1993) R3468. %--R3471
1769:
1770: \bibitem{noisy_Fred} P. Cvitanovi\'c, C.P.~Dettmann,
1771: R.~Mainieri, G. Vattay, %A. Voros,
1772: % {\em Trace formulas for stochastic evolution operators:
1773: % Weak noise perturbation theory},
1774: {\em J. Stat. Phys. \bf 93} (1998) 981;
1775: {\tt chao-dyn/9807034}.
1776:
1777: \bibitem{diag_Fred} P. Cvitanovi\'c, C.P.~Dettmann,
1778: G. Palla, N. S\o nderg\aa rd, G. Vattay,
1779: % {\em Spectrum of stochastic evolution operators:
1780: % Local matrix representation approach}
1781: {\em Phys. Rev. \bf E 60} (1999) 3936; % 3936-3941
1782: {\tt chao-dyn/9904027}.
1783:
1784: \bibitem{conjug_Fred} P. Cvitanovi\'c, C.P.~Dettmann, R.~Mainieri, G. Vattay,
1785: % {\em Trace formulas for stochastic evolution operators:
1786: % Smooth conjugation method},
1787: {\em Nonlinearity \bf 12} (1999) 939; % 939-953.
1788: {\tt chao-dyn/9811003}.
1789:
1790: \bibitem{FieldThe} P. Cvitanovi\'c,
1791: {\em Field theory}
1792: (Nordita, Copenhagen, 1983);
1793: \\
1794: {\tt www.nbi.dk/{$\sim$}predrag/field\_the/}.
1795:
1796: \bibitem{VatBS}
1797: G. Vattay and P.E. Rosenqvist,
1798: %{\em ``Periodic Orbit Quantization beyond Semiclassical Approximation"},
1799: {\em Phys. Rev. Lett. \bf 76} (1996) 335,
1800: \\
1801: {\tt chao-dyn/9509015};
1802: G. Vattay,
1803: %{\em ``Bohr Sommerfeld Quantization of Periodic Orbits"},
1804: {\em Phys. Rev. Lett. \bf 76} (1996) 1059.
1805:
1806: \bibitem{Watanabe87} S. Watanabe,
1807: %{\em ``Analysis of Wiener Functionals (Malliavin
1808: % calculus) and its applications to heat kernels}
1809: {\em Ann. of Prob. \bf 15} (1987) 1. % 1-39
1810:
1811: \bibitem{KatHass} See for example
1812: A.~Katok and B.~Hasselblatt,
1813: {\em Introduction to the Modern Theory of Dynamical Systems}
1814: (Cambridge University Press, Cambridge, 1995).
1815:
1816: \bibitem{asym_Fred} G. Palla, N. S\o ndergaard, G. Vattay, A. Voros,
1817: {\em Trace formulas for stochastic evolution operators:
1818: Beyond all orders},
1819: in preparation.
1820:
1821:
1822: %%%%%%%%%%NOT USED:
1823:
1824: \end{thebibliography}
1825:
1826: \end{document}
1827:
1828: %-------------------------------------------------------
1829: %%%%%
1830: %%%%% Below this line nothing will be printed
1831: %%%%%
1832:
1833:
1834: %%%%%
1835: %%%%%
1836: %%%%%
1837: